Ask Your Question

Revision history [back]

Amount of hugepages allocation is calculated on per-node basis using 3 controls: 2MB and 1GB HPs for nova, and memory for DPDK (in case if you use it). Then nailgun spread chosen amount over NUMA nodes.

To actually configure hugepages for compute instances:

  1. nailgun configures linux cmdline to add hugepages on boot time https://review.openstack.org/#/c/282367/
  2. puppet configures sysfs to distribute 2M hugepages over NUMA nodes https://review.openstack.org/#/c/288310/

See more on http://specs.openstack.org/openstack/fuel-specs/specs/9.0/support-hugepages.html

Amount of hugepages allocation is calculated on per-node basis using 3 controls: 2MB and 1GB HPs for nova, and memory for DPDK (in case if you use it). Then nailgun spread chosen amount over NUMA nodes.

To actually configure hugepages for compute instances:

  1. nailgun configures linux cmdline to add hugepages on boot time https://review.openstack.org/#/c/282367/
  2. puppet configures sysfs to distribute 2M hugepages over NUMA nodes https://review.openstack.org/#/c/288310/

So if you want to change the values, you can either update them via UI, or edit both /etc/default/grub and /etc/sysfs.d/hugepages.conf, then run update-grub and /etc/init.d/sysfsutils restart. 2M hugepages could be reallocated without reboot.

See more on http://specs.openstack.org/openstack/fuel-specs/specs/9.0/support-hugepages.html